Unsupported Attribution

arXiv:2605.23645 · detector citation_quote_validity · threshold_with_margin · 2026-05-25 09:50:38.647556+00:00

advisory citation_quote_validity unsupported_attribution

Citing paper attributes a specific factual claim to reference [23], which resolves to arXiv:2209.10652. The claim's distinctive tokens have only 14% overlap with any chunk of the cited paper's stored text (threshold for unsupported is 15%). The attribution could not be verified against the cited work.

Paper page Integrity report arXiv

Evidence text

Nelson Elhage, Tristan Hume, Catherine Olsson, Nicholas Schiefer, Tom Henighan, Shauna Kravec, Zac Hatfield-Dodds, Robert Lasenby, Dawn Drain, Carol Chen, et al. Toy models of superposition.arXiv preprint arXiv:2209.10652, 2022

Evidence payload

{
  "best_overlap_chunk_id": "108a9bdc-ca90-43d9-a4dd-fca5c587da2e",
  "best_overlap_score": 0.143,
  "chunks_checked": 14,
  "cited_arxiv_id": "2209.10652",
  "claim_offset": 38124,
  "claim_text": "models within the same family often share similar\nrelative token orientations, despite differing embedding dimensions; in untied Llama-3 models, they\nfind high correlations specifically in the unembedding space.",
  "ref_index": 23,
  "supported_threshold": 0.5,
  "unsupported_threshold": 0.15,
  "verdict_class": "threshold_with_margin"
}